PERSONALITY PROCESSES AND INDIVIDUAL DIFFERENCES

Disentangling the Effects of Low Self-Esteem and Stressful Events on Depression: Findings From Three Longitudinal Studies Ulrich Orth and Richard W. Robins

Laurenz L. Meier

University of California, Davis

University of Bern

Diathesis-stress models of depression suggest that low self-esteem and stressful events jointly influence the development of depressive affect. More specifically, the self-esteem buffering hypothesis states that, in the face of challenging life circumstances, individuals with low self-esteem are prone to depression because they lack sufficient coping resources, whereas those with high self-esteem are able to cope effectively and consequently avoid spiraling downward into depression. The authors used data from 3 longitudinal studies of adolescents and young adults, who were assessed 4 times over a 3-year period (Study 1; N ⫽ 359), 3 times over a 6-week period (Study 2; N ⫽ 249), and 4 times over a 6-year period (Study 3; N ⫽ 2,403). In all 3 studies, low self-esteem and stressful events independently predicted subsequent depression but did not interact in the prediction. Thus, the results did not support the self-esteem buffering hypothesis but suggest that low self-esteem and stressful events operate as independent risk factors for depression. In addition, the authors found evidence in all 3 studies that depression, but not low self-esteem, is reciprocally related to stressful events, suggesting that individuals high in depression are more inclined to subsequently experience stressful events. Keywords: self-esteem, depression, stressful events, diathesis-stress model

A growing body of research suggests that low self-esteem contributes to the development of depression. Overall, the findings support the “vulnerability model,” which states that low self-esteem operates as a risk factor for depression (Beck, 1967; Metalsky, Joiner, Hardin, & Abramson, 1993). Many previous studies used prospective designs and controlled for prior levels of both self-esteem and depression (e.g., Kernis et al., 1998; Orth, Robins, & Roberts, 2008; Orth, Robins, Trzesniewski, Maes, & Schmitt, 2008; Roberts & Monroe, 1992). The effect of low self-esteem on depression holds for men and women, from adolescence to old age, and after controlling for content overlap between self-esteem and depression scales (Orth et al., 2008, in press). Moreover, prior research has failed to support an alternative model of the relation between self-esteem and depression—the “scar model”—which hypothesizes that low self-esteem is an outcome rather than a cause of depression (Ormel, Oldehinkel, & Vollebergh, 2004; Orth et al., 2008, in press).

The Diathesis-Stress Model of Low Self-Esteem and Stressful Events Many diathesis-stress models of depression consider low selfesteem to be a predisposing factor for the development of depression (e.g., Beck, 1967; Brown & Harris, 1978; Hammen, 2005; Metalsky et al., 1993). In the face of challenging life circumstances, individuals with low self-esteem are assumed to have fewer coping resources and thus are prone to depression, whereas those with high self-esteem are assumed to have better coping resources and thus avoid spiraling downward into depression (see Hypothesis 1 in Figure 1). In other words, the experience of stressful events generally contributes to depression, but individuals with relatively high self-esteem are buffered against this effect (and, conversely, individuals with relatively low self-esteem are more vulnerable to this effect). If self-esteem buffers individuals against the deleterious consequences of stressful life events, then low self-esteem and stressful events should have an interactive effect on subsequent depression.1 The buffering hypothesis is a commonly accepted view of the causal relationship among selfesteem, stressful events, and depression. For example, Abela,

Ulrich Orth and Richard W. Robins, Department of Psychology, University of California, Davis; Laurenz L. Meier, Department of Psychology, University of Bern, Bern, Switzerland. This research was supported by Swiss National Science Foundation Grant PA001-113065 awarded to Ulrich Orth and National Institutes of Health Grant AG-022057 awarded to Richard W. Robins. We thank Nicole Messer for her help in the data collection for Study 2 and Keith Widaman and Kevin Grimm for helpful statistical advice. Correspondence concerning this article should be addressed to Ulrich Orth, who is now at the Department of Psychology, University of Bern, Muesmattstrasse 45, 3012 Bern, Switzerland. E-mail: [email protected]

1

We use the term depression to denote a continuous variable (i.e., individual differences in depressive affect) rather than a clinical category such as major depressive disorder. Taxometric analyses suggest that depression is best conceptualized as a continuous construct (Hankin, Fraley, Lahey, & Waldman, 2005; Lewinsohn, Solomon, Seeley, & Zeiss, 2000; Prisciandaro & Roberts, 2005; Ruscio & Ruscio, 2000).

Journal of Personality and Social Psychology, 2009, Vol. 97, No. 2, 307–321 © 2009 American Psychological Association 0022-3514/09/$12.00 DOI: 10.1037/a0015645

307

ORTH, ROBINS, AND MEIER

308 Hypothesis 1 Self-Esteem

Depression

Stressful Events

Hypothesis 2 Stressful Events

Depression

Self-Esteem

Depression

Self-Esteem

Hypothesis 3 Stressful Events

Hypothesis 4 Self-Esteem

Depression

Stressful Events

Hypothesis 5 Stressful Events

Depression

Self-Esteem

Figure 1. Hypotheses concerning the relations among self-esteem, stressful events, and depression. Hypothesis 1 (“self-esteem buffering”): Selfesteem buffers the effects of stressful events on depression. Hypothesis 2: Stressful events mediate the effects of low self-esteem on depression. Hypothesis 3: Low self-esteem mediates the effects of stressful events on depression. Hypothesis 4: Stressful events account for the relation between low self-esteem and depression. Hypothesis 5: Low self-esteem accounts for the relation between stressful events and depression.

Webb, Wagner, Ho, and Adams (2006) state that following negative events, “protective factors, such as high self-esteem, may prevent the outcome of depressive symptoms by decreasing the negative impact of depressogenic thoughts on the affective, cognitive, behavioral, and physiological symptoms of depression” (p. 329), and Roberts (2006), in a review on self-esteem from a clinical perspective, states that one of the themes that emerges from the depression literature is that self-esteem “interacts with other risk factors, such as life stress and attributional style, in the prediction of depression” (p. 300). The self-esteem buffering hypothesis has been tested in numerous studies, that is, whether the effects of stressful events on subsequent depression (controlling for prior levels of depression) were stronger for low versus high self-esteem individuals. Table 1 provides a summary of the findings from these previous studies. Four studies confirmed the hypothesized interaction (Abela, 2002; Fernandez, Mutran, & Reitzes, 1998; Metalsky et al., 1993; Ralph & Mineka, 1998); seven studies failed to find the hypothesized interaction (Butler, Hokanson, & Flynn, 1994; Cheng & Lam, 1997; Kernis et al., 1998; Lewinsohn, Hoberman, & Rosenbaum, 1988; Murrell, Meeks, & Walker, 1991; Roberts & Monroe, 1992; Southall & Roberts, 2002); three studies failed to find the hypothesized two-way interaction but found evidence of a three-way interaction of self-esteem, stressful events, and third variables such as dysfunctional attitudes (Abela & Skitch, 2007; Abela et al., 2006; Robinson, Garber, & Hilsman, 1995); and one study reported an interaction effect showing the opposite pattern (the effect of stressful events on depression was stronger for high vs. low self-esteem individuals; Whisman & Kwon, 1993). Thus, previous research testing the interactive effects of selfesteem and stressful events has yielded highly inconsistent results. Table 1 shows that a wide range of samples (e.g., different age groups) and designs (e.g., the prospective time intervals ranged from a few days to 2 years) were used in these studies. However, no consistent pattern emerges from these study characteristics that might help explain the divergent pattern of results. From a methodological perspective, it seems possible that the available evidence is biased by a file-drawer effect, such that studies that failed to support the buffering hypothesis were less likely to be published. Moreover, few published studies have sufficient power to detect interaction effects because (a) interactions are typically weak in magnitude (Chaplin, 1991) and (b) the interaction term in a multiple regression analysis is typically plagued by low reliability, reflecting the product of the reliability of the two variables and producing a downward bias in the interaction effect (Cohen, Cohen, West, & Aiken, 2003; Jaccard & Wan, 1995). In the present research, in contrast, we used latent variable modeling to separate construct variance from measurement error, leading to increased statistical power to test interactions and less biased estimates of the magnitude of the interaction effect (Jaccard & Wan, 1995). Moreover, we tested the self-esteem buffering hypothesis in three independent longitudinal studies.

Additional Hypotheses Concerning the Relations Among Low Self-Esteem, Stressful Events, and Depression In addition to testing the self-esteem buffering hypothesis (Hypothesis 1), we tested several additional hypotheses. Hypothesis 2

LOW SELF-ESTEEM, STRESSFUL EVENTS, AND DEPRESSION

309

Table 1 Summary of Previous Studies Testing the Interactive Effects of Self-Esteem and Stressful Events on Subsequent Depression, Controlling for Prior Depression Study

N

% female

Age group

Type of events

Prospective time interval

Self-esteem buffering interaction effecta

Abela (2002) Abela et al. (2006) Abela & Skitch (2007) Butler et al. (1994) Cheng & Lam (1997) Fernandez et al. (1998) Kernis et al. (1998) Lewinsohn et al. (1988) Metalsky et al. (1993) Murrell et al. (1991) Ralph & Mineka (1998) Roberts & Monroe (1992) Robinson et al. (1995) Southall & Roberts (2002) Whisman & Kwon (1993)

136 102 140 73 286 729 98 354 114 1,074 141 192 381 115 80

53% 86% 51% 77% 27% 52% 86% 69% — 66% 54% 64% 58% 50% 66%

adolescents adults children young adults adolescents adults young adults adults young adults adults young adults young adults adolescents adolescents young adults

discrete academic event daily hassles daily hassles stressful life events daily hassles, stressful life events stressful life events daily hassles stressful life events discrete academic event stressful life events discrete academic event discrete academic event daily hassles, stressful life eventse stressful life events daily hassles, stressful life eventse

8 weeks 6 weeksb 6 weeksb 5 months 3 months 2 years 4–5 weeks 8 months 3–7 days 2 years 10–13 days 7–9 days 4–5 months 3 months 3 months

⫹ 3-way 3-way ns nsc ⫹d ns ns ⫹ ns ⫹ ns 3-way ns ⫺

Note. Dash indicates that data were not available. “⫹” denotes a significant two-way interaction effect consistent with the self-esteem buffering hypothesis; “⫺” denotes a significant two-way interaction effect contrary to the self-esteem buffering hypothesis; “3-way” denotes a significant three-way interaction effect of self-esteem, stressful events, and a third variable, whereas the two-way interaction effect of self-esteem and stressful events was nonsignificant; “ns” denotes a nonsignificant two-way interaction effect in absence of any significant three-way interaction effect involving self-esteem and stressful events. b The data included multiple waves over 1 year, separated by 6-week intervals. Effects were estimated by multilevel modeling. c The regression equation included two-way interactions for both daily hassles and stressful life events. d Stressful life events were divided into two scales: social network events and work disruptions. The interaction of self-esteem and stressful life events was significant for social network events, but not for work disruptions. e The analyses were conducted for a variable combining daily hassles and stressful life events. a

(see Figure 1) specifies that stressful events mediate the effect of low self-esteem on depression. Specifically, individuals with low self-esteem might generate stressful events through their own behavior, which in turn contributes to depression. For example, low self-esteem is prospectively linked to aggression and other antisocial behaviors that might contribute to interpersonal conflicts and other stressful life events (Donnellan, Trzesniewski, Robins, Moffitt, & Caspi, 2005). In close relationships, low self-esteem individuals tend to perceive their partner’s behavior more negatively, thereby increasing the likelihood of relationship conflicts and rejection (Murray, Holmes, & Griffin, 2000; Murray, Rose, Bellavia, Holmes, & Kusche, 2002). We know of no previous research that has systematically tested whether low self-esteem contributes to stressful experiences. There is, however, considerable evidence that depressed individuals experience more stressful life events (Cole, Nolen-Hoeksema, Girgus, & Paul, 2006; Davila, Bradbury, Cohan, & Tochluk, 1997; Hammen, 1991; Holahan, Moos, Holahan, Brennan, & Schutte, 2005; Kim, Conger, Elder, & Lorenz, 2003). Because the link between low self-esteem and stress generation has not been tested, it is possible that depressed individuals experience more stressful events at least in part because they tend to have low self-esteem. To address this possibility, we tested whether low self-esteem and depression have independent effects on the occurrence of stressful events. If low self-esteem contributes to stressful experiences, then we can test whether stressful events mediate the effect of low self-esteem on depression. Alternatively, reversing the causal chain, low self-esteem might mediate the effects of stressful events on depression (see Hypothesis 3 in Figure 1). That is, certain stressful events (e.g., relation-

ship breakup, academic failures) may lead to low self-esteem, which, in turn, contributes to depressive affect (Oatley & Bolton, 1985; Roberts & Monroe, 1999). Although we know of no previous prospective study that has directly tested this possibility, Tram and Cole (2000) found that self-perceived competence (i.e., a construct related to self-esteem) partially mediated the effect of stressful life events on subsequent depression in a sample of adolescents. Hypotheses 4 and 5 concern possible spurious effects (see Figure 1). Hypothesis 4 specifies that stressful events lead to both low self-esteem and higher levels of depression, creating a spurious relation between the two. To test this possibility, we examined whether the relation between self-esteem and depression holds after statistically controlling for stressful events. Hypothesis 5 specifies that low self-esteem leads to stressful events and depression, creating a spurious relation between the two. In the present research, we tested whether the relation between stressful events and depression holds after controlling for the effect of self-esteem.

The Present Research All hypotheses described above can be tested in one overarching longitudinal model, which analyzes prospective main, interactive, and mediation effects. Figure 2 provides a generic illustration of this model. The self-esteem buffering hypothesis can be tested as interaction between self-esteem at one time point (e.g., Time 1) and stressful events in the subsequent time interval (e.g., Time 1–2), predicting depression at the subsequent time point (e.g., Time 2), controlling for prior depression (e.g., Time 1). The mediation hypotheses (i.e., stressful events as a mediator of the

ORTH, ROBINS, AND MEIER

310 Self-Esteem Time 1

Self-Esteem Time 2

Stressful Events Time 1-2

Depression Time 1

Self-Esteem Time 3

Self-Esteem Time 4

Stressful Events Time 2-3

Depression Time 2

Stressful Events Time 3-4

Depression Time 3

Depression Time 4

Figure 2. The figure illustrates a structural model of the relations among self-esteem, stressful events, and depression. Latent interactions are symbolized by a filled circle.

effect of low self-esteem on depression; low self-esteem as a mediator of the effect of stressful events on depression) can be tested following the guidelines outlined by Cole and Maxwell (2003). The spurious effects hypotheses (i.e., stressful events confound the relation between self-esteem and depression; self-esteem confounds the relation between stressful events and depression) can be tested by analyzing whether effects in the trivariate model differ from effects in bivariate models of the constructs. The model shown in Figure 2, and therefore all of the hypotheses, can be tested in all three longitudinal studies examined in the present research. This research extends previous studies on self-esteem, stressful events, and depression in several ways. First, in contrast to previous studies, we investigated reciprocal effects between all three constructs. By doing so, we were able to test simultaneously several models. Second, in contrast to most previous studies, we used more appropriate statistical methods on the basis of latent variable modeling, providing better estimates of the effects and more flexibility in controlling for antecedent and concurrent effects (Cole & Maxwell, 2003; Finkel, 1995). Third, in Study 1 and Study 3, we used data sets that include multiple repeated assessments (i.e., four waves), so the main and interaction effects could be replicated and constrained across multiple time points, increasing both the reliability and validity of the estimates; most previous studies used two-wave longitudinal designs, or, if more than two waves were available, most previous studies did not take full advantage of the design by testing whether effects may be constrained across time. Fourth, in Study 2 stressful events were assessed at measurement occasions that were nonoverlapping with measurement occasions for self-esteem and depression; consequently, any occasion effects (e.g., mood) that might have influenced the assessments of self-esteem and depression would be independent of occasion effects influencing reports of stressful events. Fifth, we cross-validated our results using three data sets with different sample and design characteristics; by replicating the findings across studies, we reduce methodological concerns unique to each study and strengthen confidence in the overall pattern of results. We decided to focus on adolescence and young adulthood for two reasons. First, the prevalence of depressive disorders is high during adolescence and young adulthood (Costello, Erkanli, & Angold, 2006; Kessler et al., 2005), so this developmental period

is particularly important for understanding the underlying etiology of depression. Second, self-esteem and depression are particularly likely to show changes during adolescence and young adulthood because of the many transitions that occur during this time of life (Galambos, Barker, & Krahn, 2006; Mirowsky & Kim, 2007; Robins, Trzesniewski, Tracy, Gosling, & Potter, 2002). The adolescent period is associated with rapid maturational changes, shifting societal demands, conflicting role demands, and increasingly complex romantic and peer relationships. After graduating from high school, many young adults move away from home for the first time, begin college and full-time jobs, or marry and have children. The developmental process of becoming an adult often entails a questioning of one’s identity and subsequent reformulation of conceptions and evaluations of the self. Thus, the developmental period of adolescence and young adulthood is ideally suited to test hypotheses concerning the relations among low self-esteem, stressful events, and depression.

Study 1 Study 1 used data from the Berkeley Longitudinal Study, an ongoing study of a cohort of individuals who entered the University of California, Berkeley in 1992 (for further information, see Robins, Hendin, & Trzesniewski, 2001; Robins, Noftle, Trzesniewski, & Roberts, 2005).2 We conducted six assessments over a 4-year period: first week of college; end of first semester; and end of first, second, third, and fourth years of college. We focused our analyses on the latter four assessments (denoted as Time 1–Time 4 in the remainder of this article) because stressful events and depression were not assessed in the first two assessments. To reduce age heterogeneity, we restricted the sample to participants who were 18- or 19-years-old when they entered college.

Method Participants The sample consisted of 359 individuals (59% female). Mean age of participants at Time 1 was 18.3 years (SD ⫽ 0.5, range ⫽ 2 These data were used in a previous study on the relation between self-esteem and depression (Orth et al., 2008); however, that study did not include any analyses of stressful events.

LOW SELF-ESTEEM, STRESSFUL EVENTS, AND DEPRESSION

18 –19). Of the participants, 43% were Asian, 31% Caucasian, 13% Chicano/Latino, 5% African American, 1% American Indian, 2% of other ethnicity, and 5% did not specify ethnicity. Data were available for 270 individuals at Time 1, 232 individuals at Time 2, 177 individuals at Time 3, and 277 individuals at Time 4. To investigate the potential impact of attrition, differences on study variables were tested between participants who completed the Time 4 assessment and participants who dropped out of the study before Time 4. For one variable (i.e., stressful events reported at Time 2), participants who dropped out reported higher values than participants who completed the full study (d ⫽ 0.52, p ⬍ .05). No significant differences emerged for any of the other variables (i.e., self-esteem at Times 1–3, depression at Times 1–3, and stressful events at Time 3).

Measures Self-esteem. Self-esteem was assessed with the 10-item Rosenberg Self-Esteem Scale (RSE; Rosenberg, 1965). The RSE is the most commonly used and well-validated measure of global self-esteem (Robins et al., 2001). Responses were measured on a 5-point scale ranging from 1 (not very true of me) to 5 (very true of me). The alpha reliability of the RSE was .89 at Time 1, .91 at Time 2, .90 at Time 3, and .90 at Time 4. Stressful events. At Times 2– 4, the occurrence of 12 stressful events during the past year was assessed. The items were “breaking up with boyfriend/girlfriend,” “death or serious illness/injury of a close family member or friend,” “serious personal illness or injury,” “fired from job or serious trouble with employer,” “transferred out of UC Berkeley to a different college,” “academic probation,” “victim of a crime,” “failing an important exam,” “failing a course,” “financial problems concerning school (e.g., in danger of not having sufficient money to continue),” “arrested for a crime,” and “dropped out of college.” For the analyses, the summed number of event was used, with a possible range from 0 to 12. (Coefficient alpha, which was .38 at Time 2, .30 at Time 3, and .49 at Time 4, is not an appropriate indicator of reliability because the heterogeneous items do not measure an internally consistent construct, but rather function as an index of the degree of cumulative life stress; see Bollen & Lennox, 1991; Streiner, 2003.) Depression. Depression was assessed with the Center for Epidemiologic Studies Depression Scale (CES-D; Radloff, 1977). The CES-D is a frequently used 20-item self-report measure for the assessment of depressive symptoms in nonclinical, subclinical, and clinical populations, and its validity has been repeatedly confirmed (Eaton, Smith, Ybarra, Muntaner, & Tien, 2004). Participants were instructed to assess the frequency of their reactions within the preceding 7 days. Responses were measured on a 4-point scale (0 ⫽ rarely or none of the time, less than one day; 1 ⫽ some or a little of the time, one to two days; 2 ⫽ occasionally or a moderate amount of time, three to four days; 3 ⫽ most or all of the time, five to seven days). The alpha reliability of the CES-D was .91 at Time 1, .91 at Time 2, .90 at Time 3, and .91 at Time 4.

Procedure for the Statistical Analyses The analyses were conducted using the Mplus 5.2 program (Muthe´n & Muthe´n, 2007). To deal with missing values,

311

maximum-likelihood estimation was used, which produces less biased and more reliable results compared with conventional methods of dealing with missing data such as listwise or pairwise deletion (Schafer & Graham, 2002). Models including latent interactions were estimated by numerical integration using the Monte Carlo algorithm with 2,000 integration points.3 Model fit was assessed by the Tucker-Lewis Index (TLI), the comparative fit index (CFI), and the root-mean-square-error of approximation (RMSEA), based on the recommendations of Hu and Bentler (1999) and MacCallum and Austin (2000). Hu and Bentler (1999) suggested that good fit is indicated by values greater than or equal to .95 for TLI and CFI and less than or equal to .06 for RMSEA. Because these indices are not available for models with latent interactions, the Bayesian Information Criterion (BIC) is also reported. For BIC, absolute values cannot be interpreted, but when comparing models, lower values indicate better model fit.

Results and Discussion Table 2 shows means and standard deviations of the measures used in Study 1. For the latent variables self-esteem and depression, we used item parcels as indicators because they produce more reliable latent variables than individual items (Little, Cunningham, Shahar, & Widaman, 2002). For both self-esteem and depression, we randomly aggregated the items into three parcels. In contrast, for stressful events, parceling was not possible because, as discussed above, stressful event checklists do not measure an internally consistent construct. In each study, we used the measure of stressful events as a single indicator of the latent variable and modeled the measurement error of the indicator by fixing the residual variance to 20% of the total variance of the stressful event measure (corresponding to an assumed reliability of the scale of .80).4 Before testing the study hypotheses, we first tested a basic model that did not include the interaction between self-esteem and stressful events. For self-esteem and depression, the uniquenesses of individual indicators were correlated across time to control for bias due to indicator-specific variance (cf. Cole & Maxwell, 2003). (This procedure was not appropriate for stressful events, which were measured by single indicators.) For each construct, the loading of the first indicator was set to 1 to identify the model. For self-esteem and depression, the factor loadings of the second and third indicator were constrained to be equal across time to ensure that the latent constructs were measured similarly across occasions. We accounted for variance due to specific measurement occasions by cross-sectionally correlating the disturbances of selfesteem and depression (cf. Cole & Maxwell, 2003). However, we did not include any cross-sectional correlations for stressful events 3 Given the high number of dimensions of integration in Studies 1 and 3 (i.e., six dimensions of integration), we could not use the default algorithm (i.e., rectangular integration) but used the Monte Carlo algorithm, following the recommendation by Muthe´n and Muthe´n (2007). For reasons of consistency across studies, we also used the Monte Carlo algorithm in Study 2. 4 Grant et al. (2004) reported test–retest reliability estimates for several stressful event checklists ranging from .74 to .96. We therefore decided to use .80 as a reliability estimate for the stressful event measures used in the present research.

ORTH, ROBINS, AND MEIER

312

Table 2 Means and Standard Deviations of Measures in Study 1 Time 1 (18 years)

Time 2 (19 years)

Time 3 (20 years)

Time 4 (21 years)

Variable

M

SD

M

SD

M

SD

M

SD

RSE Stressful eventsa CES-D

3.82 — 0.98

0.77 — 0.58

3.88 1.41 0.94

0.82 1.30 0.57

3.86 1.22 0.82

0.77 1.17 0.53

4.06 1.26 0.74

0.72 1.36 0.52

Note. Response scales ranged from 1 to 5 for the RSE and from 0 to 3 for the CES-D. The range of possible values for stressful events was 0 –12. RSE ⫽ Rosenberg Self-Esteem Scale; CES-D ⫽ Center for Epidemiologic Studies Depression Scale. Dashes indicate that data were not available. a Stressful events were assessed for the time interval that preceded the assessment. For example, the score at Time 2 indicates the number of stressful events between Time 1 and Time 2.

and self-esteem or for stressful events and depression because the model included direct effects between the constructs, which supersede correlated disturbances. A model constraining the stability and cross-lagged effects to be equal across the three time intervals did not lead to a significant reduction in fit relative to the unconstrained model, ⌬␹2(17, N ⫽ 359) ⫽ 9.8, ns. Consequently, we included these equality constraints in the remainder of the analyses. Table 3 shows that the overall fit of the no-interaction model was good. Table 4 shows the unstandardized estimates and standard errors of the structural coefficients. We tested two models of the interactive effect of self-esteem and stressful events on depression, one that constrained the effect to be equal across assessments and one that did not. In the constrained model, the interaction effect was close to zero and nonsignificant. In the unconstrained model, one of the three interactions was significant (see Table 4), but the other two interaction effects were in the opposite direction of the predicted effect. Moreover, the unconstrained model fit the data worse than the constrained model (as indicated by a higher value for BIC; see Table 3). Two other results shown in Table 4 are noteworthy: First, the main effects were virtually identical for the models with and without an interaction effect. Second, the standard errors were small, for both main effects and interaction effects, indicating successful model estimation. Thus, the results suggest that self-esteem and stressful events did not interact in predicting subsequent depression. Moreover, the no-interaction model fit the data better than the interaction models (as indicated by a lower value for BIC; see Table 3). We therefore examined the standardized estimates for the no-interaction model (see Figure 3). The results showed that the stability effects were significant for all three constructs and that three cross-lagged effects were significant: Both self-esteem and stressful events predicted subsequent depression, and depression predicted subsequent stressful events.5 We also tested for gender differences in the structural coefficients, using a multiple-group analysis. A model allowing for different coefficients for male and female participants did not significantly improve model fit, relative to a model with constraints across gender; the difference in model fit was ⌬␹2(13, N ⫽ 359) ⫽ 21.5, ns.6 In addition to the self-esteem buffering hypothesis, we tested several additional hypotheses. First, the effect of self-esteem on depression was not mediated by stressful events (Hypothesis 2), as indicated by a nonsignificant Sobel test. Thus, the results suggest that low self-esteem did not have a stress-generating effect. For comparison purposes, we inspected the effect sizes in a bivariate

model of self-esteem and stressful events, omitting depression. The effect of self-esteem on stressful events was small (at ⫺.05, ⫺.04, and ⫺.04 for the three time intervals, respectively; all coefficients ns). Thus, even when we analyzed the relations between self-esteem and stressful events separately from depression, low self-esteem did not predict stressful events. Second, the effect of stressful events on depression was not mediated by low self-esteem (Hypothesis 3), as indicated by a nonsignificant Sobel test. Again, for comparison purposes, we inspected the effect sizes in a bivariate model of self-esteem and stressful events, omitting depression. The effect of stressful events on self-esteem was very small and virtually at the same size as in the trivariate model (at ⫺.04 for all three time intervals; all coefficients ns).7

5

In Study 1, as well as in Studies 2 and 3, the measure of stressful events was not normally distributed. We therefore tested whether using logarithmically transformed measures of stressful events yielded different results compared with using the original scaling: In all three studies, the results were very similar and did not lead to different conclusions, in particular with regard to the interaction effects, and all of the significant paths remained significant, and the nonsignificant interaction effects remained nonsignificant. 6 In addition to testing for gender differences simultaneously for all structural coefficients, we inspected gender differences in individual structural coefficients in all three studies. There were no gender differences that replicated across studies, except that the effect of stressful events on depression was stronger for women than for men (in Study 1, the average coefficients for female and male participants were .23 and .13, respectively; in Study 2, .20 and .13, respectively; and in Study 3, .14 and .09, respectively). We also tested for ethnicity differences in the structural coefficients (Asians vs. others) in Study 1, using a multiple-group analysis. A model allowing for different coefficients for Asians versus other ethnicities did not significantly improve model fit, relative to a model with constraints across ethnicities; the difference in model fit was ⌬␹2(13, N ⫽ 359) ⫽ 18.1, ns. 7 We also tested two further mediation hypotheses. First, depression had an indirect effect on subsequent depression mediated by intermediate stressful events, as indicated by a significant Sobel test. This indirect effect was also significant in Study 3 but failed to replicate in Study 2. Thus, the temporal stability of depression is partially due to the reciprocal relationship between depression and stressful events. Second, self-esteem had an indirect effect on stressful events mediated by depression, as indicated by a significant Sobel test. This indirect effect was also significant in Study 3 (it could not be tested in Study 2 because only two waves of data were available). Thus, even if the direct effect of self-esteem on stressful events is very small, self-esteem indirectly influences the occurrence of subsequent stressful events by its effect on depression.

LOW SELF-ESTEEM, STRESSFUL EVENTS, AND DEPRESSION

313

Table 3 Fit Indices of the Models Tested in Studies 1–3 ␹2

Model Study 1 (N ⫽ 359) No interaction Unconstrained interaction Constrained interaction Study 2 (N ⫽ 249) No interaction Interaction Study 3 (N ⫽ 2,403) No interaction Unconstrained interaction Constrained interaction

ⴱ

377.2 — — 116.7 — 891.2 — —

ⴱ

ⴱ

df

TLI

CFI

RMSEA (90% CI)

BIC

290 — —

.98 — —

.98 — —

.029 (.020–.037) — —

10497.7 10519.3 10515.2

54 —

.96 —

.97 —

.070 (.053–.088) —

3964.5 3969.9

290 — —

.93 — —

.94 — —

.030 (.027–.032) — —

60573.8 60624.0 60609.2

Note. For models including interactions, ␹2, df, TLI, CFI, and RMSEA are not available. TLI ⫽ Tucker-Lewis Index; CFI ⫽ comparative fit index; RMSEA ⫽ root-mean-square-error of approximation; CI ⫽ confidence interval; BIC ⫽ Bayesian Information Criterion; dashes indicate that the fit index was not available for the model. ⴱ p ⬍ .05.

Third, the effect of self-esteem on depression held after controlling for stressful events (Hypothesis 4). In fact, the effect of self-esteem on depression was as strong as in a bivariate model of self-esteem and depression (⫺.20, ⫺.21, and ⫺.22 for the three time intervals, respectively; all ps ⬍ .05). Thus, the findings further support the influence of low self-esteem on depression, because they rule out one plausible third-variable confound— stressful life events. Fourth, the results replicate the effect of depression on stress generation (see Cole et al., 2006; Grant, Compas, Thurm, McMahon, & Gipson, 2004; Hammen, 1991) and show that this effect remains significant after controlling for self-esteem (Hypothesis 5). Thus, the findings suggest that depression and stressful events have reciprocal prospective relations—individuals experiencing stressful events are more likely to increase in depressive affect, and individuals high in depressive affect are more likely to

experience stressful events. It is interesting to note that controlling for self-esteem actually increased the size of these reciprocal effects, suggesting that self-esteem serves as a suppressor variable in this context (Paulhus, Robins, Trzesniewski, & Tracy, 2004); when self-esteem was removed from the model, the effect of stressful events on depression (at .16, .15, and .18 for the three time intervals, respectively; all ps ⬍ .05) and the effect of depression on stressful events (at .16, .14, and .15 for the three time intervals, respectively; all ps ⬍ .05) were slightly smaller. To cross-validate the findings of Study 1, we conducted two additional longitudinal studies. Study 2 allowed us to address a methodological limitation of Study 1 related to the timing of the assessment of stressful events: Although Study 1 participants were asked to report life events that occurred during the past year, they nonetheless reported those events and responded to the self-esteem and depression questionnaires in the same assessment.

Table 4 Unstandardized Estimates of Structural Coefficients in Study 1 Coefficient Effects on self-esteem Self-esteem Stressful events Depression Effects on stressful events Self-esteem Stressful events Depression Effects on depression Self-esteem Stressful events Depression IA Time 1–Time 2 IA Time 2–Time 3 IA Time 3–Time 4

No-interaction model

Unconstrained interaction model

Constrained interaction model

0.83 (0.03)ⴱ ⫺0.02 (0.01) 0.01 (0.04)

0.83 (0.04)ⴱ ⫺0.02 (0.02) 0.01 (0.05)

0.83 (0.04)ⴱ ⫺0.02 (0.02) 0.01 (0.04)

0.14 (0.11) 0.56 (0.06)ⴱ 0.41 (0.13)ⴱ

0.16 (0.13) 0.58 (0.07)ⴱ 0.42 (0.15)ⴱ

0.14 (0.12) 0.58 (0.07)ⴱ 0.41 (0.15)ⴱ

⫺0.22 (0.05)ⴱ 0.08 (0.02)ⴱ 0.28 (0.06)ⴱ — — —

⫺0.24 (0.06)ⴱ 0.08 (0.02)ⴱ 0.29 (0.08)ⴱ ⫺0.06 (0.06) 0.17 (0.08)ⴱ ⫺0.08 (0.10)

⫺0.23 (0.06)ⴱ 0.08 (0.02)ⴱ 0.28 (0.08)ⴱ 0.00 (0.05) 0.00 (0.05) 0.00 (0.05)

Note. Standard errors of estimates are given in parentheses. IA ⫽ latent interaction between self-esteem and stressful events; dashes indicate that coefficients were not estimated for that model. ⴱ p ⬍ .05.

ORTH, ROBINS, AND MEIER

314

.81*

Self-Esteem Age 18

.85*

Self-Esteem Age 19

.07

.06 -.04

-.22* Stressful Events Age 18-19

-.60* .02

-.04

-.22* Stressful Events Age 19-20

.01

.30*

Depression Age 19

-.05

-.23* Stressful Events Age 20-21

.52*

.17*

.01

.19* .18*

.20*

.20*

Self-Esteem Age 21

.06

.60*

.18*

Depression Age 18

.83*

Self-Esteem Age 20

.28*

Depression Age 20

.30*

Depression Age 21

Figure 3. Standardized structural coefficients for the “no interaction” model (Study 1). The figure shows only latent constructs and omits observed variables. ⴱ p ⬍ .05.

Study 2 In Study 2, we assessed self-esteem and depression on two occasions separated by 6 weeks (denoted as Time 1 and Time 3). In contrast to Study 1, we assessed stressful events repeatedly during the intervening period (denoted as Time 2), rather than at the same time as the assessments of self-esteem and depression. This procedure has several advantages. First, any occasion effects (e.g., mood) that might have influenced the assessments of selfesteem and depression would be independent of occasion effects influencing reports of stressful events. Second, by using multiple assessments (up to 12 in the present study) of stressful events, the reliability of the variable was significantly increased. Third, the procedure allowed us to assess a different type of stressful events: Whereas in Study 1 we examined stressful life events (some of which were major life events, such as criminal victimization or death of a close family member), in Study 2 we examined daily hassles at the workplace. Some researchers have argued that daily hassles actually have a stronger impact on psychological adjustment than major life stressors (Lazarus, 1999).

Method Participants and Procedure The sample consisted of 249 trainees (36% female) at a large Swiss company. Mean age of participants at Time 1 was 18.0 years (SD ⫽ 1.3, range ⫽ 16 –23). The data were collected using Web-based questionnaires that were accessible only to individuals who were invited to participate (individualized links were e-mailed to 270 trainees). Participants were assured that their data would be kept completely confidential. Data were available for 221 individuals at Time 1, 197 individuals at Time 2, and 185 individuals at Time 3. To investigate the potential impact of attrition, differences on study variables between participants who completed the Time 3 assessment and participants who dropped out of the study before Time 3 were tested. No significant differences emerged for any variables.

Measures Self-esteem. Self-esteem was assessed with a German version of the 10-item RSE (von Collani & Herzberg, 2003). Responses

were measured on a 6-point scale ranging from 0 (not very true of me) to 5 (very true of me). The alpha reliability was .86 at Time 1 and .89 at Time 3. Stressful events. During the first 12 workdays after the Time 1 assessment, participants were asked to report the occurrence of stressful events at the workplace. Assessments were conducted at 11.30 a.m., and participants reported events that occurred earlier in the morning. The five events were “I was criticized for my work,” “I made a mistake that will have consequences,” “I was left alone in a difficult situation,” “I had a conflict/argument with another person,” and “I was unfairly treated.” Because most of the trainees had to attend school on some of the weekdays, six daily reports were expected for each participant. However, for practical reasons, participants received e-mails providing access to the questionnaire on every weekday; therefore, the maximum number of daily reports was 12. To increase reliability, the stressful events scale was computed only if four or more daily reports were available (N ⫽ 156). The average number of daily reports was 6.1 (SD ⫽ 1.4, range ⫽ 4 –12). The number of stressful events was computed for each day, with a possible range from 0 to 5 events experienced that day, and then averaged across days. Coefficient alpha for the five events was .61; however, as noted in Study 1, coefficient alpha is not an appropriate indicator of reliability for event scales. Depression. Depression was assessed with the German 15item short form of the CES-D (Hautzinger & Bailer, 1993). Participants were instructed to assess the frequency of their reactions within the preceding 7 days. Responses were measured on a 4-point scale (0 ⫽ rarely or none of the time, 1 ⫽ sometimes, 2 ⫽ frequently, 3 ⫽ most of the time). The alpha reliability was .92 at both Time 1 and Time 3.

Procedure for the Statistical Analyses All analyses were conducted using Mplus 5.2. Model fit was assessed using the same procedures as in Study 1.

Results and Discussion Table 5 shows means and standard deviations of the measures used in Study 2. The models tested were identical to Study 1, except that the Study 2 models included only two instead of four waves of data (so that only one interaction model could be tested).

LOW SELF-ESTEEM, STRESSFUL EVENTS, AND DEPRESSION

315

Table 5 Means and Standard Deviations of Measures in Study 2 Time 1

Time 2

Time 3

Variable

M

SD

M

SD

M

SD

RSE Stressful events per day CES-D

3.78 — 0.68

0.86 — 0.56

— 0.30 —

— 0.43 —

3.74 — 0.72

0.90 — 0.59

Note. Response scales ranged from 0 to 5 for the RSE scale and from 0 to 3 for the CES-D. The range of possible values for stressful events per day was 0 –5. RSE ⫽ Rosenberg Self-Esteem Scale; CES-D ⫽ Center for Epidemiologic Studies Depression Scale; dashes indicate that data were not available.

Overall, the results replicated the findings of Study 1 (see Table 6 and Figure 4). First, we again failed to find evidence that selfesteem buffers the effects of stressful events on depression (Hypothesis 1); model fit was worse with an interaction term than without one (as indicated by a higher BIC value; see Table 3), and the interaction effect was nonsignificant. Thus, stressful events contribute to depression, regardless of whether the individual is high or low in self-esteem. Second, the effect of self-esteem on depression was not mediated by stressful events (Hypothesis 2), as indicated by a nonsignificant Sobel test. As in Study 1, the findings support the stress-generation hypothesis for depression but not for low self-esteem; moreover, we also replicated the finding from Study 1 that the effect of depression on stressful events is even stronger after controlling for self-esteem (.33 vs. .18). Third, the effect of stressful events on depression was not mediated by low self-esteem (Hypothesis 3). Fourth, the effect of self-esteem on depression remained significant after controlling for stressful events (Hypothesis 4). Fifth, the effect of depression on stressful events remained significant after controlling for self-esteem (Hypothesis 5). Finally, as in Study 1, a model allowing for different coefficients for male and female participants did not significantly improve model fit relative to a model with constraints across gender, ⌬␹2(10, N ⫽ 249) ⫽ 15.8, ns.8 Thus, Studies 1 and 2 both failed to support the interactive effect of self-esteem and stressful events on depression. However, it is crucial to evaluate whether we had sufficient statistical power to Table 6 Unstandardized Estimates of Structural Coefficients in Study 2 Coefficient Effects on self-esteem Self-esteem Stressful events Depression Effects on stressful events Self-esteem Depression Effects on depression Self-esteem Stressful events Depression IA

No-interaction model

Interaction model

0.92 (0.10)ⴱ ⫺0.21 (0.13) 0.01 (0.12)

0.93 (0.10)ⴱ ⫺0.22 (0.13) 0.02 (0.12)

0.11 (0.10) 0.23 (0.11)ⴱ

0.11 (0.09) 0.24 (0.11)ⴱ

⫺0.36 (0.11)ⴱ 0.28 (0.13)ⴱ 0.35 (0.16)ⴱ —

⫺0.37 (0.11)ⴱ 0.29 (0.13)ⴱ 0.34 (0.16)ⴱ ⫺0.04 (0.15)

Note. Standard errors of estimates are given in parentheses. IA ⫽ latent interaction between self-esteem and stressful events; dash indicates that the coefficient was not estimated for that model. ⴱ p ⬍ .05.

detect interactions, to avoid falsely accepting the null hypothesis that no interaction exists. Under the assumption that predictors are measured without error—an assumption that is acceptable in latent variable modeling because construct factors and measurement error are explicitly modeled—a sample size of at least 26 is required for large interaction effects, 55 for medium interaction effects, and 392 for small interaction effects (Cohen et al., 2003). Thus, the sample sizes of Studies 1 and 2, which were 359 and 249, respectively, were sufficiently large to detect medium but not small interaction effects. However, in psychological research, interaction effects are generally small in magnitude (Chaplin, 1991; Cohen et al., 2003). Therefore, to address potential concerns related to statistical power, we analyzed data from a very large longitudinal study in Study 3.

Study 3 In Study 3, we used data from the National Longitudinal Survey of Youth (NLSY79), a national probability survey that was started in 1979 (for further information about this study, see Center for Human Resource Research, 2006).9 The present analyses focused on the children of study participants, who were first assessed in 1994 if they had reached the age of 15 years. These adolescents and young adults were assessed biennially from 1994 to 2004, resulting in six assessments. However, the number of assessments available for each participant varies widely because there is a complex pattern of planned missing data due to budgetary reasons. For example, in 1998, only children aged 15–20 were interviewed, and in 2000, about 40% of the Black and Hispanic oversamples were not surveyed. Moreover, because at every assessment additional children reached the age of 15 years and thus became eligible for assessment, the sample size increased with every 8

Study 2 also allowed us to test whether the stability (rather than the level) of self-esteem buffers the effects of stressful events on depression, as some researchers have argued (cf. Kernis et al., 1998; Roberts & Gotlib, 1997). At the same measurement occasions as stressful events were assessed, participants completed a five-item version of the RSE, which was adapted to measure momentary self-esteem (␣ ⫽ .94). We used the intraindividual standard deviation of self-esteem across time as an indicator of (low) self-esteem stability. However, the results showed that self-esteem stability did not interact with stressful events in the prediction of subsequent depression. 9 These data were used in a previous study on the relation between self-esteem and depression (Orth et al., 2008); however, that study did not include any analyses of stressful events.

ORTH, ROBINS, AND MEIER

316 .90*

Self-Esteem Time 1

Self-Esteem Time 3

.20 -.12

-.43* Stressful Events Time 2

-.73* .01

.19* .33*

Depression Time 1

.34*

Depression Time 3

Figure 4. Standardized structural coefficients for the “no interaction” model (Study 2). The figure shows only latent constructs and omits observed variables. ⴱ p ⬍ .05.

assessment (Ns ranged from 980 in 1994 to 5,024 in 2004). The design of the study produced substantial age heterogeneity (e.g., participants in the 2004 assessment ranged in age from 15 to 34 years). To reduce the age heterogeneity of the sample, we decided to analyze sequences of four repeated assessments for those individuals who began the survey in 1994, 1996, or 1998 at the age of 15 or 16. We restructured the data for these three cohorts so that the age of every individual was 15 or 16 at Time 1.

Method

required medical attention or treatment,” and “been convicted of any charges other than a minor traffic violation.”10 For the analyses, the summed number of events was used, with a possible range from 0 to 5. Coefficient alpha was .39 at Time 2, .14 at Time 3, and .09 at Time 4; however, as noted in Study 1, coefficient alpha is not an appropriate indicator of reliability for event scales.11 Depression. As in Studies 1 and 2, depression was assessed with the CES-D. A short version of the CES-D is used in the NLSY79 with seven items: “I did not feel like eating; my appetite was poor”; “I had trouble keeping my mind on what I was doing”; “I felt depressed”; “I felt that everything I did was an effort”; “My sleep was restless”; “I felt sad”; “I could not get ‘going.’” For each item, participants were instructed to assess the frequency of their reactions within the preceding 7 days. Responses were measured on a 4-point scale (0 ⫽ rarely, none of the time, one day; 1 ⫽ some, a little of the time, one to two days; 2 ⫽ occasionally, moderate amount of the time, three to four days; 3 ⫽ most, all of the time, five to seven days). The alpha reliability of this short form of the CES-D was .65 at Time 1, .66 at Time 2, .67 at Time 3, and .68 at Time 4. (In Study 1, in which the full 20-item version of the CES-D was used, the seven-item version correlated .92 at Time 1, .91 at Time 2, .90 at Time 3, and .92 at Time 4 with the full scale.)

Procedure for the Statistical Analyses All analyses were conducted using Mplus 5.2. Model fit was assessed using the same procedures as in Study 1.

Participants The sample consisted of 2,403 individuals (50% female). Mean age of participants at the first assessment was 15.5 years (SD ⫽ 0.5, range ⫽ 15–16). Of these participants, 61% were White (non-Hispanic), 21% were Black, 12% were Hispanic, 2% were American Indian, and 4% were of other ethnicity. Data on study variables were available for 2,094 individuals at Time 1, 1,923 individuals at Time 2, 1,894 individuals at Time 3, and 2,151 individuals at Time 4. To investigate the potential impact of attrition, differences on study variables between participants who completed the Time 4 assessment and participants who dropped out of the study before Time 4 were tested. For one variable (i.e., depression reported at Time 1), participants who dropped out reported lower values than participants who completed the full study (d ⫽ ⫺0.15, p ⬍ .05). No significant differences emerged for any of the other variables (i.e., self-esteem at Times 1–3, depression at Times 2–3, and stressful events at Times 2–3).

Measures Self-esteem. As in Studies 1 and 2, self-esteem was assessed with the 10-item RSE. Responses were measured on a 4-point scale ranging from 1 (strongly disagree) to 4 (strongly agree). The alpha reliability was .84 at Time 1, .87 at Time 2, .88 at Time 3, and .88 at Time 4. Stressful events. At Times 2– 4, the occurrence of five stressful events during the past 2 years was assessed. The items were “dropped out of regular school for at least one month and then returned,” “repeated a grade in school,” “had any accidents or injuries that required medical attention,” “had any illnesses that

Results and Discussion Table 7 shows means and standard deviations of the measures used in Study 3. The models tested were identical to Studies 1 and 2. As in Study 1, we tested whether the stability and cross-lagged effects could be set equal across the three time intervals without a significant decrease in model fit. The difference in model fit of the constrained and unconstrained model was significant, with ⌬␹2(17, N ⫽ 2,403) ⫽ 70.1, p ⬍ .05. However, with sufficiently large samples, the chi-square difference test for nested models will always be significant, even when the true difference in fit is very small and theoretically irrelevant (MacCallum, Browne, & Cai, 2006). The values for RMSEA, which is less sensitive to sample size (cf. MacCallum et al., 2006), differed by only .001 between the unconstrained and constrained model. Consequently, we used constraints on all repeated structural coefficients in the remainder of the analyses. The results closely replicated the findings of Studies 1 and 2 (see Table 8 and Figure 5). Most importantly, we again failed to find evidence that self-esteem buffers the effects of stressful events on depression; model fit was worse for models with interactions than without (as indicated by higher BIC values; see Table 3), and the interaction effects were nonsignificant. Likewise, the pattern of results was similar to Studies 1 and 2 with regard to the mediation 10

The third and fourth item was assessed for the past 12 months. For Time 4, the first two items were not included in the computation of alpha reliability because the items were not applicable for most of the participants between age 19 and 21, resulting in a large number of missing data for these items at Time 4. 11

LOW SELF-ESTEEM, STRESSFUL EVENTS, AND DEPRESSION

317

Table 7 Means and Standard Deviations of Measures in Study 3 Time 1 (15 years)

Time 2 (17 years)

Time 3 (19 years)

Time 4 (21 years)

Variable

M

SD

M

SD

M

SD

M

SD

RSE Stressful eventsa CES-D

3.19 — 0.71

0.41 — 0.52

3.25 0.72 0.69

0.42 0.96 0.51

3.29 0.63 0.70

0.45 0.93 0.54

3.27 0.60 0.68

0.42 1.05 0.54

Note. Response scales ranged from 1 to 4 for the RSE Scale and from 0 to 3 for the CES-D Scale. The range of possible values for stressful events was 0 –5. RSE ⫽ Rosenberg Self-Esteem Scale; CES-D ⫽ Center for Epidemiologic Studies Depression Scale; dashes indicate that data were not available. a Stressful events were assessed for the time interval that preceded the assessment. For example, the score at Time 2 indicates the number of stressful events between Time 1 and 2.

hypotheses (Hypotheses 2 and 3) and the confounding hypotheses (Hypotheses 4 and 5). The only difference between the results of Study 3 and Studies 1 and 2 was that the effects of stressful events on self-esteem were significant— however, these effects were very small (at ⫺.03 to ⫺.04), and their statistical significance is due to the very large sample size of Study 3. Finally, as in Studies 1 and 2, we tested a model allowing for different coefficients for male and female participants. The chisquare difference was significant, with ⌬␹2(13, N ⫽ 2,403) ⫽ 31.2, p ⬍ .05; however, because of the very large sample size, we again inspected values for RMSEA, which were identical for both the unconstrained and constrained model. Consequently, we concluded that male and female participants did not differ meaningfully in the structural coefficients.

General Discussion In the present research, we investigated the main and interactive effects of low self-esteem and stressful events on depression, using three longitudinal studies of adolescents and young adults. The results from these studies provide converging support, or lack of

support, for several hypotheses concerning the relations among the three constructs. The results did not support the self-esteem buffering hypothesis, which states that the effects of stressful events (i.e., stressful life events or daily hassles) on subsequent depression are stronger for low versus high self-esteem individuals (Hammen, 2005; Metalsky et al., 1993). As reviewed in the introduction, previous empirical tests of this hypothesis have yielded inconsistent results, which may have been due to the fact that the traditional strategy of testing interaction effects (i.e., computing the product of manifest variables) suffers from unreliable coefficients and low statistical power. We bypassed these methodological problems in the present research by using a latent variable approach, and using sample sizes that provided sufficient statistical power to detect small (Study 1), medium to small (Study 2), and very small (Study 3) interaction effects. In Studies 1 and 3, statistical power was further enhanced by constraining the interaction effect across three time intervals (which reduces the standard errors of the estimates and increases the power of significance tests). Finally, the combined statistical power to detect an interaction effect in at least one out of

Table 8 Unstandardized Estimates of Structural Coefficients in Study 3 Coefficient Effects on self-esteem Self-esteem Stressful events Depression Effects on stressful events Self-esteem Stressful events Depression Effects on depression Self-esteem Stressful events Depression IA Time 1–Time 2 IA Time 2–Time 3 IA Time 3–Time 4

No-interaction model

Unconstrained interaction model

Constrained interaction model

0.61 (0.02)ⴱ ⫺0.01 (0.01)ⴱ ⫺0.03 (0.02)

0.62 (0.03)ⴱ ⫺0.01 (0.01)ⴱ ⫺0.03 (0.02)

0.62 (0.03)ⴱ ⫺0.01 (0.01)ⴱ ⫺0.03 (0.02)

⫺0.09 (0.05) 0.31 (0.02)ⴱ 0.22 (0.05)ⴱ

⫺0.09 (0.05) 0.31 (0.03)ⴱ 0.22 (0.05)ⴱ

⫺0.09 (0.05) 0.31 (0.03)ⴱ 0.22 (0.05)ⴱ

⫺0.11 (0.03)ⴱ 0.05 (0.01)ⴱ 0.50 (0.03)ⴱ — — —

⫺0.12 (0.04)ⴱ 0.05 (0.01)ⴱ 0.49 (0.04)ⴱ ⫺0.05 (0.08) ⫺0.14 (0.09) ⫺0.05 (0.08)

⫺0.12 (0.04)ⴱ 0.05 (0.01)ⴱ 0.49 (0.04)ⴱ ⫺0.07 (0.04) ⫺0.07 (0.04) ⫺0.07 (0.04)

Note. Standard errors of estimates are given in parentheses. IA ⫽ latent interaction between self-esteem and stressful events; dashes indicate that coefficients were not estimated for that model. ⴱ p ⬍ .05.

ORTH, ROBINS, AND MEIER

318

.58*

Self-Esteem Age 15

.57*

Self-Esteem Age 17

-.03

-.04 -.04*

-.08* Stressful Events Age 15-17

-.34* -.04

-.03*

-.09*

.10*

Stressful Events Age 17-19

-.03

.50*

-.04*

-.09* Stressful Events Age 19-21

.26*

.09*

-.04

.11* .10*

.11* Depression Age 17

Self-Esteem Age 21

-.03

.33*

.11* Depression Age 15

.66*

Self-Esteem Age 19

.49*

Depression Age 19

.47*

Depression Age 21

Figure 5. Standardized structural coefficients for the “no interaction” model (Study 3). The figure shows only latent constructs and omits observed variables. ⴱ p ⬍ .05.

three studies was even larger compared with the power of each individual study; however, in all three studies, the interaction effects of self-esteem and stressful events were nonsignificant. Thus, the present research provides compelling evidence that, contrary to the diathesis-stress model, low self-esteem is a risk factor for depression whose strength is not moderated by the presence or absence of stressful events and, similarly, that stressful events are a risk factor for depression whose strength does not depend on level of self-esteem. Although we failed to find support for the self-esteem buffering hypothesis, we believe that these results provide a significant contribution to the field, given the widespread belief in the validity of the hypothesis; many researchers have argued that it is critical for scientific fields to publish null results when there is a clear rationale for the hypothesis and the research is well conducted and has sufficient statistical power (Cooper, DeNeve, & Charlton, 1997; Fraley & Marks, 2007; Greenwald, 1975). The results of the three longitudinal studies also failed to confirm the hypothesis that stressful events mediate the effects of low self-esteem on depression, or that low self-esteem mediates the effects of stressful events on depression. Therefore, future research should test other mediating factors. For example, regarding the effects of low self-esteem on depression, such factors include both intrapersonal processes (e.g., low self-esteem likely elicits rumination about negative aspects of the self, which, in turn, increases and prolongs negative affect; see Nolen-Hoeksema, 2000) and interpersonal processes (e.g., low self-esteem motivates social avoidance, thereby impeding social support, which has been linked to depression; see Ottenbreit & Dobson, 2004). The findings allowed us to rule out two hypotheses concerning spurious effects. First, the effects of low self-esteem on depression were not confounded by stressful events; that is, low self-esteem and depression are not related simply because challenging life events lead to lower self-esteem and depression, creating a spurious link between the two. Second, reciprocal relations between stressful events and depression were not confounded by low selfesteem; that is, stressful events and depression are not related simply because low self-esteem individuals are prone to depression and tend to experience more stressful events, creating a spurious correlation between the two. The results of the present research provide a systematic picture of the prospective relations among the three constructs: Low

self-esteem and stressful events have independent, noninteractive effects on subsequent depression, and depression has an effect on subsequent occurrence of stressful events. The present research replicates previous studies on the relationship between stressful events and depression (cf. Cole et al., 2006; Hammen, 1991; Holahan et al., 2005) and strengthens conclusions from previous research by showing that the reciprocal relation between stressful events and depression holds after controlling for self-esteem. In addition, the present research suggests that, in contrast to depression, low self-esteem is not a factor of stress generation. It is important to note that the study design does not allow for strong conclusions regarding the causality of the effects. As in all passive observational designs, effects between factors may be caused by third variables that were not assessed (Finkel, 1995). Nevertheless, longitudinal analyses are useful because they can indicate whether the data are consistent with a causal model of the relation between the variables. Also, the results do not allow for firm conclusions with regard to the clinical category of major depressive disorder (MDD). First, the depression measures used in the present research rely on self-report, but conclusions about the antecedents of MDD should be based on diagnoses of depression from clinical interviews. Second, our analyses are based on nonclinical samples, which do not allow for valid conclusions about depressive episodes in clinical populations. Future research, therefore, should test the potential buffering effect of self-esteem on the onset or recurrence of depressive episodes in MDD following stressful events. Nevertheless, the means and standard deviations on the depression measure used (i.e., the CES-D) imply that a fairly substantial proportion of the participants in each of the three studies reported experiencing many of the depressive symptoms at least some or a little of the time or one to two days per week. Therefore, we believe that the results are relevant for levels of depressive affect that represent a significant impairment in the individual’s psychological wellbeing. Clinically significant levels of depressed mood do not necessarily have to meet the criteria for MDD as given in the Diagnostic and Statistical Manual of Mental Disorders (4th ed., text rev., American Psychiatric Association, 2000), and many persons not in treatment can have clinically significant levels of depressed mood.

LOW SELF-ESTEEM, STRESSFUL EVENTS, AND DEPRESSION

Another possible limitation of the present research is that most participants had self-esteem scores that were moderate to high in absolute value; only a small percentage of participants had scores that were below the midpoint of the response scale (specifically, 13%, 11%, and 2% in Studies 1, 2, and 3, respectively). Thus, we do not know whether the present findings generalize to samples of participants who have low self-esteem in an absolute sense. However, it is important to note that, at least in samples from Western cultures, self-esteem scores tend to be distributed predominantly in the middle to high range (cf. Heine, Lehman, Markus, & Kitayama, 1999; Robins et al., 2002). In other words, individuals who rate their self-esteem at about the midpoint of the response scale actually have low self-esteem relative to the population. It is important to note that in all three studies, the self-esteem means and standard deviations were similar to those found in normative population samples. For example, in Study 2, means and standard deviations for the German version of the RSE are similar to normative data from a representative sample, as reported by Roth, Decker, Herzberg, and Bra¨hler (2008); moreover, only about 4% of the representative sample scored below the midpoint of the response scale. Our Study 3 even provides normative data by itself, given that it uses a large probability sample. When the means and standard deviations of Study 3 are adjusted to the different response scale used in Study 1, the distributional characteristics in Study 1 correspond closely to the representative data from Study 3. Given that the distributions of self-esteem scores in our studies are similar to distributions in normative samples, the results of the present research should generalize to normal populations within the age group studied, and therefore it seems warranted to describe the findings in terms of “low” versus “high” self-esteem.12 Another limitation is that we focused on one developmental stage, specifically, adolescence and young adulthood. Future research, therefore, should test whether the results hold at other developmental stages such as midlife and old age. However, as outlined in the introduction, samples with adolescents and young adults are particularly suited to test hypotheses concerning the relations among low self-esteem, stressful events, and depression, suggesting that the present research provides a conservative test of the hypotheses. A strength of the present research is the convergence of findings across the three studies, despite different study characteristics, increasing confidence in the generalizability of the findings. The studies differed in type of sample (college students in Study 1, employee trainees in Study 2, and a national probability sample of adolescents and young adults in Study 3), nationality (American in Studies 1 and 3 and Swiss in Study 2), length of prospective time intervals (1 year in Study 1, 6 weeks in Study 2, and 2 years in Study 3), type of stressful events analyzed (stressful life events in Studies 1 and 3, and daily hassles in Study 2), and method to assess stressful events (checklists in Studies 1 and 3, and experience sampling in Study 2). Moreover, we addressed a limitation of Studies 1 and 3 in Study 2 by assessing stressful events at measurement occasions that were separate from measurement occasions for self-esteem and depression. Future studies on the self-esteem buffering hypothesis might assess the characteristics of stressful events in more detail. Selfesteem might have a buffering effect only for specific subtypes of stressful events. For example, the literature on stress generation distinguishes between so-called dependent and independent events

319

(i.e., life events that are, or are not, under the individual’s control; Hammen, 2005). In the present research, mostly dependent events were assessed: In Study 1, only one of the events was clearly independent (death or serious illness/injury of a close family member or friend), compared with none of the events in Studies 2 and 3. Therefore, in the present research, no reliable test of differences between dependent and independent events was possible. Future studies could use measures of stressful events that assess a larger set of stressful events and test the moderating effects of differing event characteristics.13 In conclusion, the present research suggests that low self-esteem and stressful events are independent risk factors for depression. These findings have implications for interventions aimed at preventing depression: Improving self-esteem reduces risk of depression regardless of whether the individual is enduring stressful or nonstressful life experiences. At the same time, preventing stressful events and improving efficacy to cope with stressful events reduces risk of depression not only among individuals with low self-esteem but also among high self-esteem individuals.

12 We examined the scatterplot of residuals (plotted against self-esteem scores) from the regression of depression on self-esteem, and they showed an essentially random distribution in all three studies, suggesting that the relation between self-esteem and depression is similar across all levels of self-esteem. To further illustrate this point, we divided the samples into self-esteem quintiles and examined mean depression scores in each quintile. These means show that depression levels declined in a fairly linear manner from the bottom self-esteem quintile (representing the 20% of participants with the lowest self-esteem levels) to the top self-esteem quintile (representing the top 20% of participants). For example, in Study 1, the mean depression scores were 1.32, 1.08, 0.87, 0.75, and 0.48 for self-esteem quintiles ordered from low to high. 13 One could argue that self-esteem only serves a buffering effect for events that are self-esteem relevant, such as academic failure. Study 1 included the most extensive measure of stressful events and therefore allowed us to test the self-esteem buffering effect specifically for academic and competence-related stressful events (5 of the 12 total events). However, no significant interaction emerged between self-esteem and this subset of events.

References Abela, J. R. Z. (2002). Depressive mood reactions to failure in the achievement domain: A test of the integration of the hopelessness and selfesteem theories of depression. Cognitive Therapy and Research, 26, 531–552. Abela, J. R. Z., & Skitch, S. A. (2007). Dysfunctional attitudes, selfesteem, and hassles: Cognitive vulnerability to depression in children of affectively ill parents. Behaviour Research and Therapy, 45, 1127–1140. Abela, J. R. Z., Webb, C. A., Wagner, C., Ho, M. H. R., & Adams, P. (2006). The role of self-criticism, dependency, and hassles in the course of depressive illness: A multiwave longitudinal study. Personality and Social Psychology Bulletin, 32, 328 –338. American Psychiatric Association. (2000). Diagnostic and statistical manual of mental disorders (4th ed., text rev.). Washington, DC: Author. Beck, A. T. (1967). Depression: Clinical, experimental, and theoretical aspects. New York: Harper & Row. Bollen, K., & Lennox, R. (1991). Conventional wisdom on measurement: A structural equation perspective. Psychological Bulletin, 110, 305–314. Brown, G. W., & Harris, T. (1978). Social origins of depression: A study of psychiatric disorder. New York: Free Press.

320

ORTH, ROBINS, AND MEIER

Butler, A. C., Hokanson, J. E., & Flynn, H. A. (1994). A comparison of self-esteem lability and low trait self-esteem as vulnerability factors for depression. Journal of Personality and Social Psychology, 66, 166 –177. Center for Human Resource Research. (2006). NLSY79 child and young adult data users guide. Columbus: Ohio State University. Chaplin, W. F. (1991). The next generation of moderator research in personality psychology. Journal of Personality, 59, 143–178. Cheng, S. K., & Lam, D. J. (1997). Relationship among life stress, problem solving, self-esteem, and dysphoria in Hong Kong adolescents: Test of a model. Journal of Social and Clinical Psychology, 16, 343–355. Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied multiple regression/correlation analysis for the behavioral sciences. Mahwah, NJ: Erlbaum. Cole, D. A., & Maxwell, S. E. (2003). Testing mediational models with longitudinal data: Questions and tips in the use of structural equation modeling. Journal of Abnormal Psychology, 112, 558 –577. Cole, D. A., Nolen-Hoeksema, S., Girgus, J., & Paul, G. (2006). Stress exposure and stress generation in child and adolescent depression: A latent trait-state-error approach to longitudinal analyses. Journal of Abnormal Psychology, 115, 40 –51. Cooper, H., DeNeve, K., & Charlton, K. (1997). Finding the missing science: The fate of studies submitted for review by a human subjects committee. Psychological Methods, 2, 447– 452. Costello, E. J., Erkanli, A., & Angold, A. (2006). Is there an epidemic of child or adolescent depression? Journal of Child Psychology and Psychiatry, 47, 1263–1271. Davila, J., Bradbury, T. N., Cohan, C. L., & Tochluk, S. (1997). Marital functioning and depressive symptoms: Evidence for a stress generation model. Journal of Personality and Social Psychology, 73, 849 – 861. Donnellan, M. B., Trzesniewski, K. H., Robins, R. W., Moffitt, T. E., & Caspi, A. (2005). Low self-esteem is related to aggression, antisocial behavior, and delinquency. Psychological Science, 16, 328 –335. Eaton, W. W., Smith, C., Ybarra, M., Muntaner, C., & Tien, A. (2004). Center for Epidemiologic Studies Depression Scale: Review and revision (CES-D and CESD-R). In M. E. Maruish (Ed.), The use of psychological testing for treatment planning and outcomes assessment: Volume 3 (Instruments for adults) (pp. 363–377). Mahwah, NJ: Erlbaum. Fernandez, M. E., Mutran, E., & Reitzes, D. C. (1998). Moderating the effects of stress on depressive symptoms. Research on Aging, 20, 163– 182. Finkel, S. E. (1995). Causal analysis with panel data. Thousand Oaks, CA: Sage. Fraley, R. C., & Marks, M. J. (2007). The null hypothesis significancetesting debate and its implications for personality research. In R. W. Robins, R. C. Fraley, & R. F. Krueger (Eds.), Handbook of research methods in personality psychology (pp. 149 –169). New York: Guilford Press. Galambos, N. L., Barker, E. T., & Krahn, H. J. (2006). Depression, self-esteem, and anger in emerging adulthood: Seven-year trajectories. Developmental Psychology, 42, 350 –365. Grant, K. E., Compas, B. E., Thurm, A. E., McMahon, S. D., & Gipson, P. Y. (2004). Stressors and child and adolescent psychopathology: Measurement issues and prospective effects. Journal of Clinical Child and Adolescent Psychology, 33, 412– 425. Greenwald, A. G. (1975). Consequences of prejudice against the null hypothesis. Psychological Bulletin, 82, 1–20. Hammen, C. (1991). Generation of stress in the course of unipolar depression. Journal of Abnormal Psychology, 100, 555–561. Hammen, C. (2005). Stress and depression. Annual Review of Clinical Psychology, 1, 293–319. Hankin, B. L., Fraley, R. C., Lahey, B. B., & Waldman, I. D. (2005). Is depression best viewed as a continuum or discrete category? A taxometric analysis of childhood and adolescent depression in a populationbased sample. Journal of Abnormal Psychology, 114, 96 –110.

Hautzinger, M., & Bailer, M. (1993). Allgemeine Depressions-Skala (ADS): Manual [Center for Epidemiologic Studies Depression Scale (CES-D): Manual]. Weinheim, Germany: Beltz. Heine, S. J., Lehman, D. R., Markus, H. R., & Kitayama, S. (1999). Is there a universal need for positive self-regard? Psychological Review, 106, 766 –794. Holahan, C. J., Moos, R. H., Holahan, C. K., Brennan, P. L., & Schutte, K. K. (2005). Stress generation, avoidance coping, and depressive symptoms: A 10-year model. Journal of Consulting and Clinical Psychology, 73, 658 – 666. Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 6, 1–55. Jaccard, J., & Wan, C. K. (1995). Measurement error in the analysis of interaction effects between continuous predictors using multiple regression: Multiple indicator and structural equation approaches. Psychological Bulletin, 117, 348 –357. Kernis, M. H., Whisenhunt, C. R., Waschull, S. B., Greenier, K. D., Berry, A. J., Herlocker, C. E., et al. (1998). Multiple facets of self-esteem and their relations to depressive symptoms. Personality and Social Psychology Bulletin, 24, 657– 668. Kessler, R. C., Berglund, P., Demler, O., Jin, R., Merikangas, K. R., & Walters, E. E. (2005). Lifetime prevalence and age-of-onset distributions of DSM-IV disorders in the National Comorbidity Survey Replication. Archives of General Psychiatry, 62, 593– 602. Kim, K. J., Conger, R. D., Elder, G. H., & Lorenz, F. O. (2003). Reciprocal influences between stressful life events and adolescent internalizing and externalizing problems. Child Development, 74, 127–143. Lazarus, R. S. (1999). Stress and emotion: A new synthesis. New York: Springer. Lewinsohn, P. M., Hoberman, H. M., & Rosenbaum, M. (1988). A prospective study of risk factors for unipolar depression. Journal of Abnormal Psychology, 97, 251–264. Lewinsohn, P. M., Solomon, A., Seeley, J. R., & Zeiss, A. (2000). Clinical implications of “subthreshold” depressive symptoms. Journal of Abnormal Psychology, 109, 345–351. Little, T. D., Cunningham, W. A., Shahar, G., & Widaman, K. F. (2002). To parcel or not to parcel: Exploring the question, weighing the merits. Structural Equation Modeling, 9, 151–173. MacCallum, R. C., & Austin, J. T. (2000). Applications of structural equation modeling in psychological research. Annual Review of Psychology, 51, 201–226. MacCallum, R. C., Browne, M. W., & Cai, L. (2006). Testing differences between nested covariance structure models: Power analysis and null hypotheses. Psychological Methods, 11, 19 –35. Metalsky, G. I., Joiner, T. E., Hardin, T. S., & Abramson, L. Y. (1993). Depressive reactions to failure in a naturalistic setting: A test of the hopelessness and self-esteem theories of depression. Journal of Abnormal Psychology, 102, 101–109. Mirowsky, J., & Kim, J. (2007). Graphing age trajectories: Vector graphs, synthetic and virtual cohort projections, and cross-sectional profiles of depression. Sociological Methods and Research, 35, 497–541. Murray, S. L., Holmes, J. G., & Griffin, D. W. (2000). Self-esteem and the quest for felt security: How perceived regard regulates attachment processes. Journal of Personality and Social Psychology, 78, 478 – 498. Murray, S. L., Rose, P., Bellavia, G. M., Holmes, J. G., & Kusche, A. G. (2002). When rejection stings: How self-esteem constrains relationshipenhancing processes. Journal of Personality and Social Psychology, 83, 556 –573. Murrell, S. A., Meeks, S., & Walker, J. (1991). Protective functions of health and self-esteem against depression in older adults facing illness or bereavement. Psychology and Aging, 6, 352–360. Muthe´n, L. K., & Muthe´n, B. O. (2007). Mplus user’s guide: Fifth edition. Los Angeles, CA: Author.

LOW SELF-ESTEEM, STRESSFUL EVENTS, AND DEPRESSION Nolen-Hoeksema, S. (2000). The role of rumination in depressive disorders and mixed anxiety/depressive symptoms. Journal of Abnormal Psychology, 109, 504 –511. Oatley, K., & Bolton, W. (1985). A social-cognitive theory of depression in reaction to life events. Psychological Review, 92, 372–388. Ormel, J., Oldehinkel, A. J., & Vollebergh, W. (2004). Vulnerability before, during, and after a major depressive episode. Archives of General Psychiatry, 61, 990 –996. Orth, U., Robins, R. W., & Roberts, B. W. (2008). Low self-esteem prospectively predicts depression in adolescence and young adulthood. Journal of Personality and Social Psychology, 95, 695–708. Orth, U., Robins, R. W., Trzesniewski, K. H., Maes, J., & Schmitt, M. (in press). Low self-esteem is a risk factor for depressive symptoms from young adulthood to old age. Journal of Abnormal Psychology. Ottenbreit, N. D., & Dobson, K. S. (2004). Avoidance and depression: The construction of the Cognitive-Behavioral Avoidance Scale. Behaviour Research and Therapy, 42, 293–313. Paulhus, D. L., Robins, R. W., Trzesniewski, K. H., & Tracy, J. L. (2004). Two replicable suppressor situations in personality research. Multivariate Behavioral Research, 39, 303–328. Prisciandaro, J. J., & Roberts, J. E. (2005). A taxometric investigation of unipolar depression in the national comorbidity survey. Journal of Abnormal Psychology, 114, 718 –728. Radloff, L. S. (1977). The CES-D Scale: A self-report depression scale for research in the general population. Applied Psychological Measurement, 1, 385– 401. Ralph, J. A., & Mineka, S. (1998). Attributional style and self-esteem: The prediction of emotional distress following a midterm exam. Journal of Abnormal Psychology, 107, 203–215. Roberts, J. E. (2006). Self-esteem from a clinical perspective. In M. H. Kernis (Ed.), Self-esteem issues and answers: A sourcebook of current perspectives (pp. 298 –305). New York: Psychology Press. Roberts, J. E., & Gotlib, I. H. (1997). Temporal variability in global self-esteem and specific self-evaluation as prospective predictors of emotional distress: Specificity in predictors and outcome. Journal of Abnormal Psychology, 106, 521–529. Roberts, J. E., & Monroe, S. M. (1992). Vulnerable self-esteem and depressive symptoms: Prospective findings comparing three alternative conceptualizations. Journal of Personality and Social Psychology, 62, 804 – 812. Roberts, J. E., & Monroe, S. M. (1999). Vulnerable self-esteem and social processes in depression: Toward an interpersonal model of self-esteem regulation. In T. Joiner & J. C. Coyne (Eds.), The interactional nature of depression: Advances in interpersonal approaches (pp. 149 –187). Washington, DC: American Psychological Association. Robins, R. W., Hendin, H. M., & Trzesniewski, K. H. (2001). Measuring global self-esteem: Construct validation of a single-item measure and the

321

Rosenberg Self-Esteem Scale. Personality and Social Psychology Bulletin, 27, 151–161. Robins, R. W., Noftle, E. E., Trzesniewski, K. H., & Roberts, B. W. (2005). Do people know how their personality has changed? Correlates of perceived and actual personality change in young adulthood. Journal of Personality, 73, 481–521. Robins, R. W., Trzesniewski, K. H., Tracy, J. L., Gosling, S. D., & Potter, J. (2002). Global self-esteem across the life span. Psychology and Aging, 17, 423– 434. Robinson, N. S., Garber, J., & Hilsman, R. (1995). Cognitions and stress: Direct and moderating effects on depressive versus externalizing symptoms during the junior high school transition. Journal of Abnormal Psychology, 104, 453– 463. Rosenberg, M. (1965). Society and adolescent self-image. Princeton, NJ: Princeton University Press. Roth, M., Decker, O., Herzberg, P. Y., & Bra¨hler, E. (2008). Dimensionality and norms of the Rosenberg Self-Esteem Scale in a German general population sample. European Journal of Psychological Assessment, 24, 190 –197. Ruscio, J., & Ruscio, A. M. (2000). Informing the continuity controversy: A taxometric analysis of depression. Journal of Abnormal Psychology, 109, 473– 487. Schafer, J. L., & Graham, J. W. (2002). Missing data: Our view of the state of the art. Psychological Methods, 7, 147–177. Southall, D., & Roberts, J. E. (2002). Attributional style and self-esteem in vulnerability to adolescent depressive symptoms following life stress: A 14-week prospective study. Cognitive Therapy and Research, 26, 563– 579. Streiner, D. L. (2003). Being inconsistent about consistency: When coefficient alpha does and doesn’t matter. Journal of Personality Assessment, 80, 217–222. Tram, J. M., & Cole, D. A. (2000). Self-perceived competence and the relation between life events and depressive symptoms in adolescence: Mediator or moderator? Journal of Abnormal Psychology, 109, 753– 760. von Collani, G., & Herzberg, P. Y. (2003). Eine revidierte Fassung der deutschsprachigen Skala zum Selbstwertgefu¨hl nach Rosenberg [A revised German version of the Rosenberg Self-Esteem Scale]. Zeitschrift fu¨r Differentielle und Diagnostische Psychologie, 24, 3–7. Whisman, M. A., & Kwon, P. (1993). Life stress and dysphoria: The role of self-esteem and hopelessness. Journal of Personality and Social Psychology, 65, 1054 –1060.

Received August 20, 2008 Revision received February 4, 2009 Accepted February 9, 2009 䡲